-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathfinetune.py
184 lines (144 loc) · 7.79 KB
/
finetune.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
import math
import torch
import logging
import numpy as np
from tqdm import tqdm
import torch.nn as nn
import multiprocessing
from os.path import join
from datetime import datetime
import torchvision.transforms as transforms
from torch.utils.data.dataloader import DataLoader
torch.backends.cudnn.benchmark= True # Provides a speedup
import homography_project
import util
import test
import dataset_geoloc
import parser
import commons
import datasets_ws
import network
import os
import warnings
warnings.filterwarnings('ignore')
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
#### Initial setup: parser, logging...
args = parser.parse_arguments()
start_time = datetime.now()
args.save_dir = join("logs_finetuning", args.save_dir, start_time.strftime('%Y-%m-%d_%H-%M-%S'))
commons.setup_logging(args.save_dir)
commons.make_deterministic(args.seed)
logging.info(f"Arguments: {args}")
logging.info(f"The outputs are being saved in {args.save_dir}")
logging.info(f"Using {torch.cuda.device_count()} GPUs and {multiprocessing.cpu_count()} CPUs")
#### Creation of Datasets
logging.debug(f"Loading dataset {args.dataset_name} from folder {args.datasets_folder}")
triplets_ds = datasets_ws.TripletsDataset(args, args.datasets_folder, args.dataset_name, "train", args.negs_num_per_query)
logging.info(f"Train query set: {triplets_ds}")
val_ds = datasets_ws.BaseDataset(args, args.datasets_folder, args.dataset_name, "val")
logging.info(f"Val set: {val_ds}")
test_ds = datasets_ws.BaseDataset(args, args.datasets_folder, args.dataset_name, "test")
logging.info(f"Test set: {test_ds}")
############### MODEL ###############
features_extractor = network.FeaturesExtractor(args)
args.features_dim = global_features_dim = 384#*64
if args.resume_fe != None:
# state_dict = torch.load(args.resume_fe)
# features_extractor.load_state_dict(state_dict)
state_dict = torch.load(args.resume_fe)["model_state_dict"]
from collections import OrderedDict
new_state_dict = OrderedDict()
for k, v in state_dict.items():
k = k.replace("module.backbone","encoder")
name = k.replace("module.aggregation","pool")
new_state_dict[name] = v
features_extractor.load_state_dict(new_state_dict)
else:
logging.warning("WARNING: --resume_fe is set to None, meaning that the "
"Feature Extractor is not initialized!")
homography_regression = network.HomographyRegression()
if args.resume_hr != None:
state_dict = torch.load(args.resume_hr)
homography_regression.load_state_dict(state_dict)
model = network.Network(features_extractor, homography_regression).cuda().eval()
model = torch.nn.DataParallel(model)
### Setup Optimizer and Loss
if args.optim == "adam":
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)
elif args.optim == "sgd":
optimizer = torch.optim.SGD(model.parameters(), lr=args.lr, momentum=0.9, weight_decay=0.001)
criterion_triplet = nn.TripletMarginLoss(margin=args.margin, p=2, reduction="sum")
best_r5 = start_epoch_num = not_improved_num = 0
#### Training loop
for epoch_num in range(start_epoch_num, args.epochs_num):
logging.info(f"Start training epoch: {epoch_num:02d}")
epoch_start_time = datetime.now()
epoch_losses = np.zeros((0,1), dtype=np.float32)
# How many loops should an epoch last (default is 5000/1000=5)
loops_num = math.ceil(args.queries_per_epoch / args.cache_refresh_rate)
for loop_num in range(loops_num):
logging.debug(f"Cache: {loop_num} / {loops_num}")
# Compute triplets to use in the triplet loss
triplets_ds.is_inference = True
triplets_ds.compute_triplets(args, model)
triplets_ds.is_inference = False
triplets_dl = DataLoader(dataset=triplets_ds, num_workers=args.num_workers,
batch_size=args.train_batch_size,
collate_fn=datasets_ws.collate_fn,
pin_memory=(args.device=="cuda"),
drop_last=True)
model = model.train()
# images shape: (train_batch_size*4)*3*H*W
# triplets_local_indexes shape: (train_batch_size*2)*3 ; because 2 triplets per query
for images, triplets_local_indexes, _ in tqdm(triplets_dl, ncols=100):
# Flip all triplets or none
if args.horizontal_flip:
images = transforms.RandomHorizontalFlip()(images)
# Compute features of all images (images contains queries, positives and negatives)
features, features_a = model("features_extractor", [images.to(args.device), "global"])
loss_triplet = 0
if args.criterion == "triplet":
triplets_local_indexes = torch.transpose(
triplets_local_indexes.view(args.train_batch_size, args.negs_num_per_query, 3), 1, 0)
for triplets in triplets_local_indexes:
queries_indexes, positives_indexes, negatives_indexes = triplets.T
loss_triplet += criterion_triplet(features[queries_indexes],
features[positives_indexes],
features[negatives_indexes])
random_weights = (torch.rand(args.train_batch_size, 4)**0.1).cuda()
REIloss = homography_project.reprojection_error_ofinliers(model, features_a[queries_indexes], features_a[positives_indexes], weights=random_weights)
loss_REI = torch.sum(REIloss)
del features
del features_a
loss_triplet /= (args.train_batch_size * args.negs_num_per_query)
loss_REI /= args.train_batch_size
loss_joint = loss_triplet + 100*loss_REI
optimizer.zero_grad()
loss_joint.backward()
optimizer.step()
batch_loss = loss_joint.item()
epoch_losses = np.append(epoch_losses, batch_loss)
del loss_joint
print("triplet loss:",loss_triplet.item())
print("REI loss:",loss_REI.item())
logging.debug(f"Epoch[{epoch_num:02d}]({loop_num}/{loops_num}): " +
f"current batch joint loss = {batch_loss:.4f}, " +
f"average epoch joint loss = {epoch_losses.mean():.4f}")
logging.info(f"Finished epoch {epoch_num:02d} in {str(datetime.now() - epoch_start_time)[:-7]}, "
f"average epoch joint loss = {epoch_losses.mean():.4f}")
torch.save(homography_regression.state_dict(),
f"{args.save_dir}/homography_regression_{epoch_num:03d}.torch")
torch.save(features_extractor.state_dict(),
f"{args.save_dir}/features_extractor_{epoch_num:03d}.torch")
############### TEST ###############
geoloc_test_dataset = dataset_geoloc.GeolocDataset(args.datasets_folder, args.dataset_name, split="val",
positive_dist_threshold=args.val_positive_dist_threshold)
logging.info(f"Geoloc test set: {geoloc_test_dataset}")
logging.info(f"now it's test time")
test_baseline_recalls, test_baseline_recalls_pretty_str, test_baseline_predictions, _, _, query_img, gallery_img = \
util.compute_features(geoloc_test_dataset, model, global_features_dim)
logging.info(f"baseline test: {test_baseline_recalls_pretty_str}")
if args.dataset_name =="msls" or (args.dataset_name =="pitts30k" and epoch_num>=35):
_, reranked_test_recalls_pretty_str = test.test(model, test_baseline_predictions, query_img, gallery_img, geoloc_test_dataset,
num_reranked_predictions=args.num_reranked_preds, recall_values=[1,5,10,20])
logging.info(f"test after re-ranking - {reranked_test_recalls_pretty_str}")